A methodological framework to explore latent travel patterns and estimate typical OD matrices: A case study using Brisbane Bluetooth multi-density OD database

Limited studies exist in the literature on OD related travel patterns, and the analysis of same demands rich database of OD matrices with appropriate clustering algorithms. The OD matrices representing different travel patterns are generally multi-density high dimensional data points. To identify different clusters, we need appropriate proximity measures and clustering algorithms. This paper develops a comprehensive methodological framework to explore latent travel patterns from multi-density high dimensional OD matrices estimate typical OD matrices corresponding to those patterns. The main takeaways of the study are as follows: first to cluster high-dimensional OD matrices we deploy geographical window based structural similarity index (GSSI) as structural proximity measure in the DBSCAN algorithm; second to address the issue of multi-density data points, we propose clustering on individual subspaces; third, we propose a simple two-level approach to identify optimum DBSCAN parameters, and finally, we identify OD matrix clusters, typical travel patterns and estimate typical OD matrices for the study region. The proposed framework is applied on real Bluetooth based OD (B-OD) matrices from Brisbane City Council (BCC) region and we have found nine typical OD matrices corresponding to nine typical travel patterns. Although the study demonstrated the application using B-OD matrices it is applicable for OD matrices developed from other data sources such as Mobile Phone and smartcard etc. The knowledge of travel patterns and typical OD matrices has many practical applications such as assisting transport modellers in simulating realistic traffic conditions and policy makers in making effective and rational policy developments.